Court Rules Pentagon Can Blacklist Anthropic for Refusing to Enable Claude Features
A federal court has ruled that the Department of Defense holds legal authority to blacklist Anthropic from government contracting if the AI company declines to enable specific features in its Claude models for military applications. The decision establishes a significant precedent for how federal agencies can condition AI procurement on developer compliance with capability requirements — even when those requirements conflict with a company's stated safety policies.
The ruling arrives at a moment when the U.S. government is accelerating its integration of commercial AI systems across defense and intelligence operations. It directly tests whether AI developers retain the authority to restrict how their models behave when deployed in federal contexts, or whether procurement leverage can compel capability access that companies would otherwise refuse.
At issue is whether Anthropic's safety-based restrictions on Claude — limitations the company characterizes as deliberate design choices to prevent harm — constitute grounds for exclusion from federal contracts, or whether the government's demand to override those restrictions crosses a legal or constitutional line. The court concluded the government's position was permissible: conditioning contract eligibility on capability compliance is within the executive branch's procurement authority.
The practical mechanics of the dispute center on specific Claude features that the Pentagon reportedly sought to enable for defense use cases — capabilities Anthropic had restricted as part of its responsible deployment framework. Anthropic's refusal to unlock those features, the court found, gave the DoD sufficient basis to treat the company as non-compliant with contract terms, and therefore eligible for exclusion from future awards.
The implications extend well beyond Anthropic. Every major AI developer with federal contracts — or aspirations to hold them — now faces a sharper version of a tension that has been building for years: the conflict between internal safety governance and the operational demands of government clients. A company's acceptable use policy, previously treated as a product decision, now carries direct procurement consequences when the customer is the federal government.
For AI developers that have built safety restrictions into their models as non-negotiable constraints, the ruling signals that those constraints may not be legally protected when they conflict with government requirements. The government is not being ordered to accept restricted tools — it is being confirmed as having the right to reject them.
For defense contractors and systems integrators working to embed commercial AI into military infrastructure, this ruling clarifies the risk profile of building on models from developers with restrictive deployment policies. Dependence on a vendor that may be blacklisted introduces supply chain fragility that procurement officers and program managers will now need to account for explicitly.
The longer-term dynamic this creates is a bifurcation in the AI market. Developers willing to negotiate capability access with the federal government — or to offer separate deployment configurations for defense use — will hold a structural advantage in competing for contracts. Developers that treat safety restrictions as uniform and non-negotiable across all customers will face increasing exclusion from the federal sector, which remains one of the largest and most stable sources of AI infrastructure investment.
What this ruling does not resolve is whether the government's demanded capabilities would expose Claude to uses that create downstream legal or ethical liability, or whether the precedent will encourage agencies to push capability demands further in future procurement cycles. Those questions are likely to generate additional litigation. What is now settled is the baseline: the government can close its doors to AI vendors who decline to meet its configuration requirements, and courts will not intervene to prevent that.
Sources: — Ars Technica (https://arstechnica.com/tech-policy/2026/09/court-rules-trump-can-blacklist-anthropic-for-refusing-to-enable-claude-features/)